Abstract
This work describes
a model (called Model D)
for optimizing the TPC-D queries
developed
under the Cascades Optimizer Framework.
Cascades
[Graefe 95]
is an extensible framework
for building
rule-based, top-down
database query optimizers.
Model D
is limited
to a subset of SQL
sufficient for representing the TPC-D queries,
with a very simple
physical model
(no parallelism or buffer management),
a limited number of physical operators
and a limited search space.
The 17 TPC-D queries
were hand parsed
into a logical algebra query tree
and submitted to the optimizer.
This work demonstrates
the extensibility
of the Cascades framework.
It also demonstrates
that the both the development
of good transformation rules
and the selection and computation
of logical and physical properties
to achieve accurate cost models
continue to be a black art.
But absent a sufficiently static
data model environment,
such an extensible optimizer framework
could be better than
traditional bottom-up optimizers.
