Distributed vs local
A distributed system and a single local system differ in two major ways:
- The local system can rely on a loopback interface for internal communication. A distributed system requires communication between nodes using a common network.
- The local system has unhindered access to the local files. A distributed system requires to use a shared network drive with higher latency.
So, there is additional work involved to set the simulation up.
Common network interface
To setup communication, preCICE needs to exchange the endpoint information between participants.
sockets, preCICE defaults to endpoints of the loopback interface, which only works on the local system.
Follow the instructions on how to configure the communication to select an appropriate network.
The common working directory requires a common filesystem.
Most cluster administrators provide a shared filesystem by default. It is used to share datasets and provide common software to all nodes in the system. preCICE does not require a fast underlying filesystem, but high latency on file creation can noticeably slow the initialization down.
In case you don’t have access to a managed cluster and you want to use available workstations, then you have to setup this shared filesystem yourself.
The simplest may be to use a common network file share using samba. This is available on virtually all Linux distributions.
Another option is to mount a remote directory using an SSH Filesystem (
In any case, make sure that both are available as the same path on both machines. You can use symbolic links to simplify this task.
Running the simulation
Now open a terminal on each machine and navigate to the test setup. Then run your solvers.
If your cluster is managed using SLURM, then further steps may be necessary.