Selection of object surrogates to support clustering

摘要

A new object identification scheme for object-oriented databases is suggested. Its purpose is to choose object surrogates so that they convey information about preferred clustering of stored objects. The surrogates are composed of two fields: cluster code and sequence number. A tailored variant of extendible hashing is proposed for the access method. A hash function is evaluated to produce the cluster code at object creation time, so that surrogates themselves act as hashed pseudokeys. The problem how to manage variable-size logical clusters is solved by chopping the large ones into physical subclusters of restricted size. The suggested approach shares the advantages of plain surrogates and structured addresses: it is a logical identification scheme, but the indirection of references is implemented effectively. Experiments confirm the benefits of the proposed organization.