Querying DAG-shaped Execution Traces Through Views

The question whether a given set of views, defined by queries, can be used to answer another query, arises in several contexts such as query optimization, data integration and semantic caching [24, 10, 12]. This paper studies a specific instance of this problem, where the queried data has the shape of a DAG (Directed Acyclic Graph) and the query language uses DAG patterns to retrieve portions of the data graph that are of interest. Our study is motivated by a particular application domain concerning the analysis of Web-based Business Processes (BPs for short). Such DAGs / DAG patterns are the standard way to model / query BP execution traces [3]. Previous research considered tree-shaped XML data and (general) graph-shaped Semi-Structured data. We show that the particular DAG shape of BP execution traces makes the problem easier than for general graphs, yet harder than for XML trees. Specifically, we show which combinations of DAG classes and query features allow for PTIME query answering algorithms and which lead to NP-complete problems.