Building CCA Tools on the Cray XT

Many aspects of HPC systems at Leadership Class Facilities (LCF) require special configurations for building the CCA Tools. This page includes example commands and handy downloadable shell scripts for configuring and building Babel, the CCA Specification, Ccaffeine (Ccafe), and Bocca on the Cray XT systems at ORNL (Jaguar) and NERSC (Franklin).


Note: Each of the scripts linked from this page will require _modification_ to set the filesystem paths and directories where the CCA Tools' source code repositories and installed binaries should go. This organization is based on a typical installation for LCF machines, where the source code for the CCA Tools and user application codes is stored in the user's _home area_, whereas the actual binaries/libraries for execution on the back-end compute nodes must be stored in a _distinct_ set of directories on the LCF system's _scratch area_ (which is the _only_ filesystem visible from the compute nodes for actual production runs!). Aside from the compute node accessibility requirements, each filesystem area typically has its own disk usage and/or inode quotas, and you can't run any jobs if you surpass your quotas, so putting everything together on one filesystem probably won't work. :-)

The key variables that you will need to modify are:

• "top" = the top level _home area_ directory where the CCA Tools source code directories are stored.

• "scratch" = the main directory in the _scratch area_ where the installed binaries and libraries should be installed.

Remember, you must properly set these variables in each of the scripts below. For consistency (and hopefully clarity), the remainder of this documentation also refers to these variables as a generic means for indicating various filesystem directories.

Btw, if you don't want to mess with the scripts/settings, then please just ignore them and manually type the commands below. :-D


Because the LCF machines are invariably "bleeding edge" systems, the standard "stable" CCA Tools tar balls and contractor builds do not (as yet) work "out of the box" for these systems. That may change, as the special instructions on this wiki page are incorporated into the CCA contractor build specification. But for now, the suggested build approach is to check out the source code for the CCA Tools straight from the various repositories, requiring appropriate login accounts on "cca-forum.org" and "sourceforge.org".


Because file storage space is typically limited on LCF machines, a VBuild structure is used in the instructions here (i.e. in most cases we don't build the CCA Tools in the same directory where the source was checked out, but instead configure and build the tools over on the scratch filesystem). Further, because LCF machines regularly clean, whack, delete, rebuild and otherwise purge their scratch areas, we recommend storing any/all of these scripts in some nice directory in your _home area_, which is hopefully more stable (and backed up :-)) and then linking the scripts into the VBuild directories where you configure and build stuff.


To check out the latest version of the Babel subversion repository:


Next, you must configure Babel for use on the _compute_ nodes of the Cray XT.
This can be done using the provided doconf.babel.compute script.

For the script-urally challenged, these are the effective (csh) commands that the script executes:


This should give you a functional Babel installation.


Blah blah blah...


Blah blah blah...


Blah blah blah...