OPTIMAL SELECTION OF SECONDARY INDICES IN RELATIONAL DATABASES