openmpi-kma

Change log

Image on Docker Hub - https://hub.docker.com/r/olehkurpiak/openapi-kma

1.2

added host file. gives ability to simulate more processors, then available on machine. #1

1.1

Initial release

Docker

Run Docker image using docker run -it -v PROJECT_LOCATION:/tmp/project -v LOCAL_M2_PATH:/root/.m2 --name openmpi-kma olehkurpiak/openapi-kma:1.1

PROJECT_LOCATION - path to java project on your local machine

LOCAL_M2_PATH - path to yours .m2 folder(usually under ~/.m2) - this will allow to cache maven dependencies, otherwise maven would require to download them on each new container start

File mpi.jar is located under path /tmp/openmpi/lib/mpi.jar inside container. You can extract it to your local machine using command docker cp openmpi-kma:/tmp/openmpi/lib/mpi.jar MY_LOCAL_PATH/mpi.jar. Then you can add it to IDEA project as a jar dependency

Maven

You can also build java project using Maven with-in docker container.

Add following dependency to you maven project

<dependency>
    <groupId>mympi</groupId>
    <artifactId>mympi</artifactId>
    <version>1.0</version>
    <scope>system</scope>
    <systemPath>/tmp/openmpi/lib/mpi.jar</systemPath>
</dependency>

Go to cd /tmp/project inside container (Note that this is a path in container where project was mounted from local machine. This is from first command)
Run mvn clean install / mvn clean compile

Running "Hello world" with-in container

Inside container run command mpirun -np 1 java -cp /tmp/project/target/classes com/mathpar/NAUKMA/examples/HelloWorldParallel (Note: /tmp/project as a root of classpath)

Hostfile

Initially docker could run only 5 processors but it can be increased. The file which sets the number of available processors is located under path /root/hostfile inside a container. By default number of processors is set to 40.

Running "Hello world" with-in container with hostfile

Inside container run command mpirun --hostfile /root/hostfile -np 20 java -cp /tmp/project/target/classes com/mathpar/NAUKMA/examples/HelloWorldParallel (Note: /tmp/project as a root of classpath)

borsch / openmpi-kma