Umeå universitets logga

Businesses and organization relies on process querying to analyse data using tools such as process querying languages (PQLs) or SQL. As PQLs introduce certain drawbacks and as SQL remains a standard for handling data, accessing SQLs usefulness for process querying is of practical relevance. The most common ways to express such queries in SQL are with MATCH_RECOGNIZE, and regular expression (REGEXP) based SQL approaches. However, the usefulness of MATCH_RECOGNIZE is still unknown given that SQL have already well-established REGEXP support, and many SQL engines not yet supporting MATCH_RECOGNIZE.

In this thesis we empirically evaluate the performance and scalability off MATCH_RECOGNIZE in comparison to REGEXP-based SQL approaches for process querying by using a simple dataset derived from SIGNAL—a PQL by SAP Signavio—and translating the SIGNAL patterns to both MATCH_RECOGNIZE and REGEXP-based SQL queries. The execution time, CPU usage, and peak-memory usage is measured for each query, and to evaluate the scalability the dataset size is varied using logarithmic scaling e.g., 10%, 25%, 50%,75%, 100% for each query.

The findings of the experiment showed that REGEXP-based SQL approaches outperform MATCH_RECOGNIZE in all metrics, often by a factor of 2. The results did also find that both approaches does scale linearly with increasing data size.

These findings indicate that MATCH_RECOGNIZE might not be the best tool for all process querying task in SQL, especially when using a simple dataset. However, we strongly speculate that MATCH_RECOGNIZE does outperform REGEXP-based SQL approaches as the complexity of the dataset increases.