Ubuntu+open mpi+ssh Small cluster simple configuration __parallel

In order to complete the course assignments, the author is committed to find the most simple configuration methods. The goal is to connect three nodes to run a simple MPI program.

Reference: https://www.open-mpi.org, https://blog.csdn.net/kongxx/article/details/52227572 each node initialization

To facilitate subsequent SSH connections, set all nodes to have the same username (because SSH IP is equivalent to SSH $USER @ip). Please refer to https://blog.csdn.net/silent56_th/article/details/80033636 for the section on creating a new user.

An optional operation to facilitate subsequent input is to map a name to the IP address of the other node by modifying/etc/hosts in each node for reference. password-free SSH connection between nodes

Detailed reference: https://blog.csdn.net/kongxx/article/details/47046817

In a nutshell, Node A uses an algorithm such as RSA to generate a public key for identity authentication, and Node B authorizes node A by recording the public key of Node A. After authorization is complete, node A can access Node B without a password.

Theoretically only need other nodes to authorize the main node, the author in order to prevent bugs, all of the nodes are authorized. Introduction and compilation of Open MPI

For a brief introduction to MPI and open MPI, see: Https://www.open-mpi.org/video/?category=general

Simply to say that MPI is the protocol, Open MPI is a concrete implementation.

About the compilation of open MPI: Https://www.open-mpi.org/faq/?category=building#easy-build

Easy steps: Download decompression –>configure (--prefix for specified installation location) –> sudo make all install

The author directly installs the open MPI to the system, so the--prefix=/user/local, and the subsequent need to set the path and other system parameters, or the appropriate location of the bin and Lib path into the $PATH and $LD _library_path (you can . BASHRC settings) compiling and running code

The MPICC command is similar to the GCC/CC usage, and the simple operation is MPICC <c_filename>-o <excutable_filename>

Single node execution: MPIRUN-NP <processor_num> <filename>, where-NP specifies the number of processors, and the first unexplained string (<filename>) is considered an executable file name.

Small cluster implementation: MPIRUN-NP <processor_num>--hostfile

localhost   slots=2
slave1  slots=2
slave2  slots=2

Where localhost is the local node, SLAVE1/2 is the IP address of the other two nodes in the primary node (set in/etc/hosts)