Scientific Data Management: The Quest for More Abstraction
Kai Lin
San Diego Supercomputer Center

SQL is the time-tested, "high level", ad hoc query language for user data
interactions which supports the querying (and modification) of structured
databases. In scientific applications, the data are much more complex, e.g.
2D and 3D geospatial data, spatiotemporal data, data streams, acyclic and
cyclic graphs, 3D molecular and protein structures, etc. Furthermore, users
prefer to interact with the data at much higher levels of abstractions that
are consistent with the respective scientific domain's view of data, rather
than a database view of data.

This presentation will describe some on-going projects at the San Diego
Supercomputer Center, such as the Geosciences Network (GEON,
www.geongrid.org), which are developing data management systems for
scientific applications. Next, we will describe user requirements that have
arisen in such projects which require us to address this notion of a "higher
abstraction" for data interactions. We will describe the types of operations
that the data management system needs to support in order to provide users
such an abstraction, and therefore a "good user experience". There are many
alternative strategies that need to be examined and evaluated, and many open
issues spanning both physical and logical database design. The primary
objective of this talk is to present these open challenges.