Authors:
Marcel Weisgut
;
Leonardo Hübscher
;
Oliver Nordemann
and
Rainer Schlosser
Affiliation:
Hasso Plattner Institute, University of Potsdam, Potsdam, Germany
Keyword(s):
Resource Allocation Problems, Stochastic Workloads, Index Selection, Robustness, Linear Programming.
Abstract:
Fast processing of database queries is a primary goal of database systems. Indexes are a crucial means for the physical design to reduce the execution times of database queries significantly. Therefore, it is of great interest to determine an efficient selection of indexes for a database management system (DBMS). However, as indexes cause additional memory consumption and the storage capacity of databases is limited, index selection problems are highly challenging. In this paper, we consider a basic index selection problem and address additional features, such as (i) multiple potential workloads, (ii) different risk-averse objectives, (iii) multi-index configurations, and (iv) reconfiguration costs. For the different problem extensions, we propose specific model formulations, which can be solved efficiently using solver-based solution techniques. The applicability of our concepts is demonstrated using reproducible synthetic datasets.