Speaker
Dan Flath
(SLAC)
Description
The Data Handling Pipeline ("Pipeline") has been developed for the Gamma-Ray Large
Area Space Telescope (GLAST) launching at the end of 2007. Its goal is to generically
process graphs of dependent tasks, maintaining a full record of its state, history
and data products. In cataloging the relationship between data, analysis results,
software versions, as well as statistics (memory usage, cpu usage) of the processing
it is able to track the complete provenance of all the data products.
The pipeline will be used to automatically process the data down-linked from the
satellite and to deliver science products to the GLAST collaboration and the Science
Support Center. It is currently used to perform Monte Carlo simulations, and
analysis of commissioning data from the instrument. It will be stress tested this
summer with "end-to-end" tests of data processing from the satellite and a full 1
year simulation run.
The Pipeline software is written almost entirely in Java and comprises several
modules. A set of Java Stored Procedures compiled into the Oracle database allow
computations on data to occur without network overhead. The Pipeline Server module
accepts user requests, performs remote job scheduling and submission, and processes
small "scriptlets" that allow lightweight calculations without the overhead of a
batch job. The Pipeline Server submits jobs to the SLAC batch farm (3000+ linux
cores), but will soon also submit jobs to a batch farm in France, and via the Grid to
a farm in Italy. The "Pipeline Front End" displays live processing statistics via
the web. It also provides AIDA charts summarizing CPU and memory usage, average
submission wait time and also provides a graphical work-flow representation of the
processing logic. Pipeline administrators can interact with the pipeline via web
based or line-mode clients.
| Submitted on behalf of Collaboration (ex, BaBar, ATLAS) | GLAST |
|---|
Author
Dan Flath
(SLAC)
