Losslessly Stripping Non-Applicable Nulls From a Database

Abstract

SQL nulls have been criticized in the literature for lacking formal specifications in the standard, which created confusion around the interpretation of null values in databases. A recent survey studied the variations in null interpretations between database practitioners and found that the majority interpret the null value as a non-applicable null, which is when the value doesn’t exist. The authors clarified that research on this area is limited and suggested that upcoming work focuses non-applicable nulls. Thus, our thesis test the applicability of the relational schema decomposition theory, mentioned in the survey, and study the performance of the decomposed database in practice. We develop the tool in Python and perform experimental evaluation on three databases to measure the memory size and query runtime and compute the overhead as our evaluation metrics. The main findings are that the performance inversely related to the number of nullable columns and that the decomposed database perform worse than the same database before decomposition.