Experience with representing c++ program information in an object-oriented database

Two major issues related to storing program information in an OODB are sharing and clustering. The former is important since it prevents the database from consuming excessive disk space, while the latter is crucial, since it keeps clients running without thrashing. In our database, objects are shared across multiple programs' translation units, and are clustered by combining three techniques, namely, birth-order, death-order, and sharing-oriented clusterings. An initial experiment shows that, for a medium-size application, the database consumes 3.5 times less disk space than in a conventional environment, and that the invocation of a client is almost instantaneous. © 1994, ACM. All rights reserved.