A transaction-oriented approach to attribute partitioning

摘要

This study introduces the concepts of sufficient and support to design two methods for solving the attribute partitioning problem: MAX and FORWARD SELECTION. These concepts express the simple idea that in order to minimize costs, a segment should contain all the attributes required by some transaction(s) and exclude other attributes. Contrary to all previous studies which without exception treat the attribute as the decision variable and may become computationally infeasible when the number of attributes is large (a condition favoring attribute partitioning), these two methods treat the transaction as the decision variable and their run time is not affected by the number of attributes. Both methods generate excellent results without using complicated mathematics. They also take into account the complexity caused by the interaction between attribute partitioning and access path selection, an issue most previous studies fail to address. Besides, both methods provide detailed information normally not available from other methods. This information provides new insight into the process of attribute partitioning. MAX is for cases involving a modest number of transactions (⩽15), FORWARD SELECTION for cases involving a large number of transactions.