Abstract for CHEP98 for the Storage Management component

[Arie Shoshani <arie@math.lbl.gov>]

This was NOT submitted yet.  Any comments are welcome.
------

Storage Management for High Energy Physics analysis applications

L. Bernardo, H. Nordberg, D. Rotem, A. Shoshani, A. Sim

National Energy Research Scientific Computing (NERSC) Division 
Lawrence Berkeley National Laboratory
Berkeley, California

We describe the design, architecture and progress of the Storage
Management component of the High Energy and Nuclear Physics Grand
Challenge (HENP-GC) project at LBNL.  This component addresses the
problem of efficient organization of the event data stored on
tertiary storage in order to  minimize the number of tape mounts and
files accessed when subsets of the events are requested.  The design
is based on partitioning the events into cells defined over bins in
the property space of the events.  Several aspects of the system are
described in detail:  the design and implementation of a clustering
analyzer (which analyzes clustering properties of the events in a
high-dimensional space), bit-sliced indexing methods over the binning
space, support for caching policies of files for multiple, possibly
overlapping, queries, and a cache manager which interacts with the
HPSS mass storage management system.