Use of large databases in scientific research has increased to the
point where scientists can test hypotheses, perform research, and
make original discoveries entirely within the database.  Several
projects in medicine, biology, biochemistry, physics and astronomy
provide very large repositories (multi-terabyte) of raw data for
other scientists to then carry out experiments ``inside'' the
database.  The Sloan Digital Sky Survey``SkyServer'' is one example
of this, with 14TB of image data and spectroscopic analyses freely
available for access by any internet user.  Many of these projects,
as a matter of principle, offer extremely liberal access policies:
SkyServer, for example, accepts and executes arbitrary SQL queries
from any remote user without authentication.  The only constraints
are some fairly generous runtime timeouts, where queries are
abandoned after 10min--1hr if they have not completed.

We outline one possible contribution of programming language
technology to managing resource and availability issues in this
model: specifically, a type and effect system for a nested
relational query language that provides asymptotic complexity bounds
for query execution time and result size.  The goal is that such
bounds can form the basis of useful feedback to users about
potentially expensive queries, as well as a portable and transparent
policy language for resource management in globally-accessible
scientific databases.

This talk describes work done at the University of Edinburgh in
collaboration with James Cheney and Kenneth MacKenzie.