Running on Big Red II

Harlan runs on IU's new supercomputer, Big Red II, but there are some hoops to jump through. Here's how to do it.

1. Install Petite. You'll want to install this in your home directory and make sure it's in your path. I use the threaded, 64-bit Linux version of Petite, but the non-threaded one should work too. Steps that worked at least once were:

• Change to csvxxx/custom
• ./configure --installprefix=~ --installman=~/.man
• Edit the Mf-install and change InstallOwner to be yourself and InstallGroup to be a group you belong to (such as csci).
• make
• make install
• Will create /bin/petite...which should be executable if /bin is on your path

2. Clone the Harlan repo. You'll probably need to register a new set of keys for Big Red II. I've had trouble cloning over HTTP, but SSH works. If you're having problems with git...try module load git to change to the most current version of git
3. Launch an interactive job on a GPU node: qsub -I -l nodes=1 -q gpu This will kick you back up to your home directory, so you'll need to change back to the Harlan directory.
4. Load the latest git: module load git (yes...again). For some reason the default version is 1.6 something, which doesn't handle HTTPs on GitHub.
5. Run aprun make
6. To actually run anything that uses the GPU, you'll need to prefix the command with aprun. For example, to build and run the test suite, do aprun make check and then aprun test.bin/ioat.kfc.bin (or any other xxx.kfc.bin in test.bin)