Speaker
Jerome LAURET
(BROOKHAVEN NATIONAL LABORATORY)
Description
While many success stories can be told as a product of the Grid middleware
developments, most of the existing systems relying on workflow and job execution are
based on integration of self-contained production systems interfacing with a given
scheduling component or portal, or directly uses the base component of the Grid
middleware (globus-job-run, globus-job-submit). However, such systems usually do not
take advantage of the presence of Resource Manager System (RMS); they hardly allow
for a mix of local RMS and are either Grid or non-grid enabled. We intend to present
an approach taking advantage of both worlds.
The STAR Unified Meta-Scheduler (SUMS) project provides users a way to submit jobs on
a farm, at a site (multiple pools or farms) or on the Grid without the need to know
or adapt to the diversity of technologies and knowledge involved while using multiple
LRMS and their specificities. The strategy was adopted in 2002 to shield the users
against changes in technologies inherent to the emerging Grid infrastructure and
developments.
Java based and taking as input a simple user job description language (U-JDL), SUMS
allows connection with multiple (overlapping or not) LRMS and Grid job submission
(Condor-G, grid-job-submit, …) without the need for changing the U-JDL. Fully
integrated with the STAR File and Replica Catalog, information providers (load and
queue information), SUMS provides a single point of reference for users to migrate
from a traditional to a distributed computing environment. Results and the
evolutionist architecture of the SUMS will be presented and its future, improvements
and evolution will be discussed.
Primary authors
D. Olson
(Lawrence Berkeley Laboratory)
E. Efstathiadis
(BROOKHAVEN NATIONAL LABORATORY)
G. CARCASSI
(BROOKHAVEN NATIONAL LABORATORY)
I. Sakrejda
(NERSC)
Jerome LAURET
(BROOKHAVEN NATIONAL LABORATORY)
L. Didenko
(BROOKHAVEN NATIONAL LABORATORY)
V. Fine
(BROOKHAVEN NATIONAL LABORATORY)
