The Language Technology Group (LTG) at the University of Oslo (UiO) is about to start a joint preparatory project with the UiO Research Computing Services group (VD), aiming to scale up DELPH-IN technology for efficient use on high-performance computing (HPC) resources. Please see the full grant application, if you require more detailed background information. A more high-level overview of DELPH-IN-related language technology research at UiO is available as slides copies from a presentation held at the 2008 annual NOTUR meeting. In 2008 and 2009, the project is funded by the Norwegian Metacenter for Computational Science (NOTUR).

This page, at least initially (June 2009), mostly serves project-internal purposes. It is intended to provide the necessary background information on DELPH-IN software and typical use patterns for VD staff, on the one hand, and to document emerging recipes for using DELPH-IN tools on the UiO TITAN facilities, on the other hand.

The page was created and is maintained by StephanOepen. Please be conservative in making (non-trivial) changes, i.e. please make sure you are reasonably sure of the information you post here.

There is an archived mailing list for this project: titan@delph-in.net. All project members should be subscribed already, but please feel free to contact StephanOepen in case you feel you should be added to the list. Ideally, all email relevant to the project should be sent to the mailing list.

The project was granted a resource alloation of 10,000 cpu hours on TITAN for the allocation period 2009.1 (April 1 through September 30, 2009). This allocation is available as account number nn9106k, which should be used in all SLURM (the queueing system on TITAN) commands. To gain access to these cpu allocations, user accounts need to be registered with the project; please contact StephanOepen for details.

The project will use the so-called LOGON tree for its experimentation, essentially an integrated, ready-to-run distribution of core DELPH-IN components. Please see the LogonTop and LogonInstallation pages for basic instructions on obtaining a complete installation, but please read the following paragraphs first.

Because the project will depend on a few add-on components in the LOGON tree, notably the full Allegro Common Lisp environment, all project members will require a personal DELPH-IN SVN user account, and always use the authenticated SVN access method (through http://logon.emmtee.net/ rather than through the world-readable, unauthenticated http://svn.emmtee.net/); please see the LogonExtras page for details.

To get started, run the following command (on a suitably modern Un*x system):

  htpasswd -n $USER

Please choose a low-security (i.e. web-quality) password and email the output of the command to StephanOepen. This information will determine your DELPH-IN SVN user account.

Once your SVN account is active, as a quick-start guide, try the following:

  svn co http://logon.emmtee.net/trunk logon
  cd logon/franz
  svn switch http://logon.emmtee.net/extras/trunk/franz
  cp ~oe/src/logon/franz/linux.x86.32/devel.lic linux.x86.32
  cp ~oe/src/logon/franz/linux.x86.64/devel.lic linux.x86.64

Next follow the instructions on the LogonInstallation page, completing at least the remaining parts of step (1) (i.e. the first-time account setup, adding a few LOGON-specific settings to your personal shell configuration file .bashrc). Once these steps are complete (including logging out and back in, for the changes to take effect), confirm basic functionality by running a lightweight and short job:

  $LOGONROOT/parse --erg mrs

In some cases, the LOGON tree only includes pre-compiled binaries for software that the project is likely to modify, specifically PET and TADM. Source code for PET and TADM is hosted in separate SVN repositories, and we will likely create private branches for use in this project.

At present, the DELPH-IN tools are not immediately applicable to large-scale HPC environments. This is in part owed to missing 'packaging' and interfaces (for effective batch operation), and in part owed to a lack of support for distributed and parallel processing. The main objectives of the project are to (a) adapt the core DELPH-IN tools for effective use on the UiO TITAN cluster and (b) to train the UiO members of DELPH-IN in the daily routines of HPC usage. The project is organized as four sub-tasks, each with its own wiki page for specific information: (1) Packaging, Scripting, and Training (TitanPackaging); (2) MPI Support in TADM (TitanTadm); (3) Profiling and Optimizing PET (TitanPet); and (4) MPI Support in [incr tsdb()] (TitanItsdb).