Often a job's workflow has different requirements at different
stages of processing.  It would be wasteful in a massively parallel
system to collect hundreds of processors, only to make them wait
while the data files are copied or other serial tasks are
performed.  By the same token,  if only a minimal number of
processors are requested for the bulk of the job's processing, it
would be nice to increase the number of processors for the parallel
calculations that could benefit. In LoadLeveler, this type of
workflow adjustment is called "staging" or "stepping."

Staging is accomplished in the LoadLeveler (LL) submit script.  A
typical LL script has a keyword stanza, followed by the keyword
"queue" and then a shell command execution space.  If, a single
stanza exists in the LL script,  the job is always assigned
"step 0."  This can be confirmed by noticing that job IDs always
seem to end in a zero. However, multiple stanzas can be linked
together in a single LL script to form a workflow.

Suppose we want to run a job that will:

1.  Copy a [previously compiled executable] file and a data file
      to a [previously created] scratch directory,
2.  Run the executable on 48 processors,
3.  Copy the results to a safe place, and
      remove the output files from the scratch directory.

This can be completely accomplished in a single script and the job
need never reserve more processors than is required for that portion
of the workflow.

How do we link the stanzas together?  We introduce two new keywords,
step_name and dependency. We will first name each stanza with
step_name, and then use the dependency keyword to require a
successful completion of the previous step prior to executing the
next one. Each step is allocated its own time block. So, if a queue
limit is one week and six job steps are used, the resulting workflow
could run as long as six weeks. There is virtually no limit to the
number of steps which may be strung together. One ocean model job
required nearly one hundred! A simple example of the case presented
above may be found at:

     http://rac.uits.iu.edu/hpc/loadleveler_ex.shtml

Use of Research Technologies systems requires a communications
client known as the secure shell (ssh). Graphical applications
require another package known as an X server. These software
packages can be a hassle to install and configure. For convenience,
Research Technologies has created a CD that provides both the secure
shell and an X server that are run directly from the CD without the
hassle of installing or configuring software. The CD is known as
XLiveCD, and it is available from http://xlivecd.indiana.edu/.

After you download a copy and burn it to CD, here's how to use it:

1. Insert CD into the drive
2. Accept the license and indicate that you want to run the software
3. Indicate the number of buttons your mouse has
4. Wait for a window to appear

When the window appears, you can click in it and then run the
secure shell to connect to a system. For example, user hoagyc
would use the following command to connect to Big Red:

   ssh hoagyc@bigred.teragrid.iu.edu

You'll then be asked to accept the host's key of authenticity and
to supply your password. You can then run applications, and
graphical applications will create windows on your workstation.

XLiveCD does not yet run on Windows Vista. Briefly, the software
that was used to build XLiveCD does not yet support Windows Vista.
We are watching the development of that software, and we will
release a new version of XLiveCD when the software is stable.

XLiveCD has been very popular. Since December 2004 ,over 1000 copies
have been downloaded by people at IU, and over 175,000 copies have
been downloaded throughout the world.

Wednesday, July 25, 12:30-1:30 -- ICTC Room 497 & IMU Walnut Room:
Research Technologies Round Table

Scott McCaulay will discuss IU's role in the TeraGrid.

--------

Tuesday, July 10 -- Submission deadline

Workshop on Progress Toward Petascale Applications in Bioinformatics
and Computational Biology, to be held in conjunction with the IEEE
7th International Symposium on Bioinformatics & Bioengineering
(BIBE 2007), which will be held at Cambridge-Boston, Massachusetts,
USA, October 14-17, 2007.

Paper Submission Final Deadline has been extended to July 10, 2007.
All accepted paper will be published by IEEE indexed in EI, INSPEC,
DBLP and Library of Congress, and can be further included in journal
issues dedicated for IEEE 7th BIBE such as BMC with unique PubMed ID
for each paper (both SCI and PubMed/Medline indexed).

--------

Sunday, October 14, 2007 -- Boston, MA

Indiana University is offering a TeraGrid-related tutorial, "Using
IU's Big Red PowerPC Cluster and IU Storage Resources via the
TeraGrid" at BiBE 2007 (Bioinformatics and Biomedical Engineering),

The primary purpose of this tutorial is to enable TeraGrid users to
learn about the Big Red system so that they can easily use codes
already ported and optimized for that system (e.g. WRF, NAMD, MILC),
or rapidly migrate other applications to Big Red.

In addition, as massive computations commonly depend on massive data
sets as input, and produce massive data sets as output, it may be
useful to obtain a working knowledge of IU's archival data storage
system, and how to store and access files via gridftp.

Plan to attend to gain hands-on experience with Big Red and IU's
High Performance Storage System, as resources on the TeraGrid. For
more information, see

  http://www.cs.gsu.edu/BIBE07/index.php