Poster: Twins, a Middleware for Adaptive Streaming Provenance at the Edge

Paper i proceeding, 2021

Data streaming applications process continuous flows of data to detect unusual/critical events. When it is beneficial to further analyze the source data leading to such events, fine-grained streaming provenance can be used to link each event back to its contributing data. Existing provenance tools, though, (i) can be computationally heavy, especially for applications deployed on resource-constrained devices at the edge of Cyber-Physical Systems, and (ii) cannot activate/deactivate provenance recording based on user-defined rules. To cover such gaps, we present Twins, a new adaptive provenance tool that leverages APIs found in state-of-the-art streaming frameworks to allow for custom conditions to activate/deactivate provenance recording. Our preliminary results, based on an implementation on top of Apache Flink and GeneaLog show that Twins can match, during the periods in which provenance is inactive, the performance of queries that do not record provenance at all.