Speaker
David Adams
(BNL)
Description
DIAL is a generic framework for distributed analysis. The heart of the system is a
scheduler (also called analysis service) that receives high-level processing requests
expressed in terms of an input dataset and a transformation to act on that dataset.
The scheduler splits the dataset, applies the transformation to each subdataset to
produce a new subdataset, and then merges these to produce the overall output dataset
which is made available to the caller. DIAL defines a job interface that makes it
possible for schedulers to connect with a wide range of batch and grid workload
management systems. It also provides command line, root, python and web clients for
job submission that enable users to submit and monitor jobs in a uniform manner.
Scaling to very large jobs can be handled with a scheduler that does partial
splitting and submits each subjob to another scheduler. I will give the current
status of DIAL and discuss its use in the context of the ATLAS experiment at the CERN
LHC (Large Hadron Collider). There we are looking at submission to local batch
systems, globus gatekeepers, EGEE/LCG workload management, ATLAS production, and
PANDA. The latter is a U.S. ATLAS framework for data production and distributed
analysis (thus the name) that may also use DIAL for its internal scheduling.
Primary author
David Adams
(BNL)
Co-authors
Chun Lik Tan
(University of Birmingham)
Karl Harrison
(University of Cambridge)
