Speaker
Mr
Lars Schley
(University Dortmund, IRF-IT, Germany)
Description
This paper discusses an architectural approach to enhance job scheduling in data
intensive applications in HEP computing. First, a brief introduction to the current
grid system based on LCG/gLite is given, current bottlenecks are identified and
possible extensions to the system are described. We will propose an extended
scheduling architecture, which adds a scheduling framework on top of existing
compute and storage elements. Goal is an improved and better coordination between
data management and workload management. This includes more precise planning and
prediction of file availability prior to job allocation to compute elements, as
well as better integration of local job and data scheduling to improve response
times and through-put. Subsequently, the underlying components are presented, where
for the design of the computing element standard grid components are used. The
storage element is based on the dCache software package that supports a scalable
storage and data access solution, which is enhanced in a way that it can interact
with scheduling services. For broader acceptance of the scheduling solution in Grid
communities beyond High Energy Physics, an outlook is given on how the scheduling
framework can be adapted to other application scenarios like e.g. the climate
community.
The project is funded by the German Ministry of Education and Science as part of
the national e-science initiative D-Grid and is jointly carried out by IRF-IT of
University of Dortmund and DESY.
Primary authors
Mr
Alexander Papaspyrou
(University Dortmund, IRF-IT, Germany)
Mr
Lars Schley
(University Dortmund, IRF-IT, Germany)
Mr
Martin Radicke
(DESY, Hamburg, Germany)
Dr
Michael Ernst
(DESY, Hamburg, Germany)
Dr
Patrick Fuhrmann
(DESY, Hamburg, Germany)
Dr
Ramin Yahyapour
(University Dortmund, IRF-IT, Germany)
