Midway examples

Quick start

A Hello World example:

$ nice ./test-login.sh hello.swift

The PR-1 workflow:

$ nice ./test-login.sh pr-1.swift

File index

test-login.sh

Runs workflow on login nodes

test-compute.sh

Runs workflow on login nodes. See settings.sh.

settings.sh

Swift/T PATH settings and SLURM job settings (if needed).

hello.swift

Hello world.

py-1.swift

Simple Python case.

py-loop-1.swift

Simple Python loop case.

Parareal workflows

Based on Figure 2 in https://doi.org/10.1145/2132876.2132883

pr-1.swift

Initial idea - simplified dependency structure

Compute nodes

To run on the compute nodes:

1. Edit settings.sh to set your PROJECT (account) and QUEUE (partition). These variable names are known to Swift/T and are used by it on all schedulers.

2. Run test-compute.sh, which provides -m slurm to swift-t.

3. Swift/T will:

   1. Create and report the name of a new directory for the job, called TURBINE_OUTPUT. This behavior is customizable. This is the PWD for the job.

   2. Create and report the name of a new turbine-slurm.sh script, which it then submits. It does not delete this script, so you can inspect it for debugging.

   3. Run the job. You can check on it using the normal SLURM tools.

   4. Redirect output to TURBINE_OUTPUT/output.txt.

4. When the job is done, check output.txt for output, timing, and debugging information.