Matrix Clustering Algorithms for Vertical Partitioning Problem: an Initial Performance Study

Matrix clustering algorithms are among the oldest approaches to the vertical partitioning problem. They can be summarized as follows: (1) given a workload, construct an Attribute Usage Matrix (AUM), (2) apply some kind of a row and column permutation algorithm and (3) extract the resulting clusters which define the required fragments. This naive approach holds some promise for a number of contemporary applications: (1) dynamization of vertical partitioning (2) big data applications and other cases of resource constraints (3) tuning of multistores. In this paper we examine a number of existing matrix clustering algorithms used for vertical partitioning. We study these algorithms and assess the quality of the solutions. The experiments are run on the TPC-H workload using the PostgreSQL DBMS.