Knowledge Representation in the Large

Abstract

Frame knowledge representation systems lack two important capabilities that prevent them from scaling up to large applications: they do not support fast access to large knowledge bases (KBs), nor do they provide concurrent multiuser access to shared KBs. We describe the design and implementation of a storage subsystem that submerges a database management system (DBMS) within a knowledge representation system. The storage subsystem incrementally loads referenced frames from the DBMS, and can save to the DBMS only those frames that have been updated in a given session. We present experimental results that show our approach to be an improvement over the use of flat files, and that evaluate several variations of our approach.