Abstract
The topic for this thesis is index management in Database management
Systems (DBMS).
The thesis is split in two related parts. The first part documents the
design and implementation of Rindex, a platform whose goal is to
support research on index management in RAM. This work (both the
programing and the writing) was done in collaboration with Tomas Are
Haavet.
The idea behind the project is to keep all indexes but the primary
ones entirely in main memory, and to create indexes on all attributes
whose value domain is strictly ordered. Another fundamental concept is
that instead of applying the query operators to relations, we apply
them to indexes. In it's current implementation, Rindex is only
concerned with queries, and only a subset of the query language from
SQL is implemented.
The second part first establishes a methodology for evaluating Rindex,
and then uses this methodology for an actual evaluation. Here Rindex
is evaluated by doing a comparative analysis against Oracle. The
analysis is done by measuring the execution time of the two platforms
for a set of selected queries. The queries are partitioned into five
test cases, that delves into the basic operations: joins, selections
and projections, but also index generation.
The platform is far too immature, so an absolute conclusion can not be
done. However the tests all indicate that there might be significant
performance gains by managing indexes the ``Rindex way''.