Search
Task to practice on seminar: EN: 06_seminar_assignment-en.pdf CZ: 06_cviceni_zadani.pdf
Open-MPI specification and documentation
Wes Kendall's tutorial
Program for task 2 (test program):
#include <stdio.h> #include "mpi.h" int main(int argc, char *argv[]) { int numprocs, rank, namelen; char processor_name[MPI_MAX_PROCESSOR_NAME]; MPI_Init(&argc, &argv); MPI_Comm_size(MPI_COMM_WORLD, &numprocs); MPI_Comm_rank(MPI_COMM_WORLD, &rank); MPI_Get_processor_name(processor_name, &namelen); printf("Hello from process %d out of %d on %s\n", rank, numprocs, processor_name); MPI_Finalize(); return 0; }
Compilation:
mpiCC program.c -lm -o program
mpirun
1. Generate ssh key
/usr/bin/ssh-keygen -N "" -t ecdsa -f ~/.ssh/id_ecdsa cd ~/.ssh/ cat id_ecdsa.pub >> authorized_keys chmod -c 600 authorized_keys
or simpler variant
ssh-keygen ssh-copy-id login@localhost
ssh-copy-id ensures that default public key file ~/.ssh/id_rsa.pub or ~/.ssh/id_ecdsa content is copied to ~/.ssh/authorized_keys file on the target computer. In our case, the copy target is SSH on same computer as we are logged in. You need to login by your password. It is not necessary to copy public key because all computers in the laboratory mounts your home directory from same NFS volume export. ABut your home directory is mounted on each computer after first login of given user. This means that first login requires password authentization. That is why you should keep one remote shell connection to each computer which should be automatically used by MPI. Laboratory computer addresses are from range 192.168.202.101 to 192.168.202.120.
ssh-copy-id
~/.ssh/id_rsa.pub
~/.ssh/id_ecdsa
~/.ssh/authorized_keys
192.168.202.101
192.168.202.120
2. Connect through ssh to all computers/nodes inteded for your program execution
3. Run
mpirun --host 192.168.202.101,192.168.202.102 -np 2 ./program
To suppress hidden infinite wait for target system signature by SSH use
mpirun --mca plm_rsh_args "-oStrictHostKeyChecking=no"
How to find IP address of computer where is given process of MPI program run: You can use the following code to determine the IP address of the computer where program is run. Thanks to Martinov Velekov for providing solution. Remark: Pointer returned bygetenv() use only to immediately read string pointed by it. Next getenv() call can change content of memory location returned by previous call.
#include <string.h> char* remote_ip = getenv("SSH_CONNECTION"); char localhost[] = "localhost"; char * pch; if(NULL != remote_ip) { pch=strchr(remote_ip,' '); pch=strchr(++pch,' '); remote_ip = ++pch; } else remote_ip = localhost;
NetBeans setup for programs compilation:
C Compiler: - Include directories: .../Program Files/MPICH2/include Linker: - Additional Library directories: .../Program Files/MPICH2/lib - Libraries: mpi.lib
program.c compilation from Cygwin command line “Cygwin Bash Shell” (without NetBeans use) using g++ (gcc, mpiCC):
g++ program.c -I"C:\Program Files\MPICH2\include" -L"C:\Program Files\MPICH2\lib" -lmpi
mpiexec.exe wrapper is used to run MPICH2 programs. smpd.exe manager has to be run/available on each computer(it is MPICH2 processes manager). Program mpiexec.exe can be controlled from command line or from interactive wmpiexec.exe console. Following figure shows example how to run program.
Availability and names of available neighborhood computers can be obtained from wmpiconfig.exe as is shown on the next figure (entries emphasized by green color are available). Click to Get Hosts first and then to Scan Hosts.
Remark: Do not forget check availability or copy required libraries to the executable file directory (usual file names: cyggcc_s-1.dll, cyggomp-1.dll, cygstdc++-6.dll, cygwin1.dll).