Institutional Repository
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Distributed query monitoring through convex analysis: towards composable safe zones

Garofalakis Minos, Samoladas Vasilis

Full record


URI: http://purl.tuc.gr/dl/dias/D2084A51-F673-469B-BFF1-657F14BF7E8C
Year 2017
Type of Item Conference Full Paper
License
Details
Appears in Collections

Summary

Continuous tracking of complex data analytics queries over high-speed distributed streams is becoming increasingly important. Query tracking can be reduced to continuous monitoring of a condition over the global stream. Communication-efficient monitoring relies on locally processing stream data at the sites where it is generated, by deriving site-local conditions which collectively guarantee the global condition. Recently proposed geometric techniques offer a generic approach for splitting an arbitrary global condition into local geometric monitoring constraints (known as "Safe Zones"); still, their application to various problem domains has so far been based on heuristics and lacking a principled, compositional methodology. In this paper, we present the first known formal results on the difficult problem of effective Safe Zone (SZ) design for complex query monitoring over distributed streams. Exploiting tools from convex analysis, our approach relies on an algebraic representation of SZs which allows us to: (1) Formally define the notion of a "good" SZ for distributed monitoring problems; and, most importantly, (2) Tackle and solve the important problem of systematically composing SZs for monitored conditions expressed as Boolean formulas over simpler conditions (for which SZs are known); furthermore, we prove that, under broad assumptions, the composed SZ is good if the component SZs are good. Our results are, therefore, a first step towards a principled compositional solution to SZ design for distributed query monitoring. Finally, we discuss a number of important applications for our SZ design algorithms, also demonstrating how earlier geometric techniques can be seen as special cases of our framework.

Services

Statistics