Year: 2007
Type: article
Abstract: Data lineage plays a critical role in verifying data correctness in scientific databases and data warehousing.In this paper, we clearly define forward data lineage in bag semantics and show its properties.We propose a tracing method that piggybacks normal query evaluation.Our method effectively supports aggregation and variable granularity lineage.More importantly, it features cost effective tracing through sophisticated implementation.The implementation, based on roBDDs, exploits the clustering and overlapping characteristics of data lineage to significantly reduce the overhead.Prior lineage tracing techniques compute lineage backward.More specifically, given a query that produces a view, the lineage is computed by executing the reverse query.While such techniques are believed to be very efficient due to the lazy computation, they are indeed inferior to the proposed forward method in most aspects.value t ⇒ DL(t j , v) = {t j }, j ∈ {1, ..., n} according to the lineage definition ⇒ Any lineage set that produces value t must contains one of the DL(t j , v) according to the relational semantics.Let t 1 , ..., t n be the set of result tuples that have value t ⇒ DL(t j , v) =< {t j1 }, ..., {t jm } >, j ∈ {1, ..., n} ⇒ Any lineage set that produces value t must be the superset of one of the DL(t j , v) according to the relational semantics.(2Let t 1 , ..., t n be the set of result tuples that have value t ⇒ DL(t j , v k ) = DL(t j , w <k ), j ∈ {1, ..., n} -(i) ⇒ Assume there is a DL x that does not subsume any DL(t j , v k ) but v k ( DL x ) = {t} ⇒ DL x does not subsume any DL(t {1,...,n} , w <k ) and w <k ( DL x ) = {t} ⇒ Contradiction to the induction assumption.The remaining cases can be proved similarly.
Authors: Mingwu Zhang, Xiangyu Zhang, Sunil Prabhakar
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W46188810