Data stream fusion for accurate quantile tracking and analysis

UDDSketch is a recent algorithm for accurate tracking of quantiles in data streams, derived from the DDSketch algorithm. provides accuracy guarantees covering full range independently input distribution and greatly improves with regard to DDSketch. In this paper we show how compress fuse two or more streams (or datasets) by leveraging mergeability summaries. general, summaries on are said be mergeable if there exists an that allows combining into single one related union datasets, simultaneously preserving error size guarantees. The property sketch enables parallel distributed processing big volume can compressed fused means such structures. Among applications strictly quantiles, requiring and/or recall here estimating latency web site, database query optimizers need succinctly summarizing values occurring over sensor network. We prove fully introduce PUDDSketch, version suitable message-passing based architectures. formally its correctness compare it DDSketch, showing through extensive experimental results our almost always outperforms overall determining quantiles. Moreover, also design implement versions both state art KLL REQ sequential algorithms order contrast PUDDSketch versus corresponding algorithms. Our experiments clearly faster par running time, whilst providing greater accuracy.