Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Functional geometric monitoring for distributed streams

Samoladas Vasilis, Garofalakis Minos

Full record


URI: http://purl.tuc.gr/dl/dias/26827170-47CA-4DD5-9AE2-B52E1CB2C8EC
Year 2019
Type of Item Conference Full Paper
License
Details
Bibliographic Citation V. Samoladas and M. Garofalakis, "Functional geometric monitoring for distributed streams," in 22nd International Conference on Extending Database Technology, 2019, pp. 85-96. doi: 10.5441/002/edbt.2019.09 https://doi.org/10.5441/002/edbt.2019.09
Appears in Collections

Summary

We introduce Functional Geometric Monitoring (FGM), a substantial theoretical and practical improvement on the core ideas of Geometric Monitoring. Instead of a binary constraint, each site is provided with a non-linear function, which, applied to its local summary vector, projects it to a real number. The sites collectively monitor the sum of these one-dimensional projections and as long as the global sum is subzero, the monitoring bounds are guaranteed. We demonstrate that FGM is as generally applicable as Geometric Monitoring, and provides substantial benefits in terms of performance, scalability, and robustness. In addition, in FGM it is possible to prove worst-case results, under standard monotonicity assumptions on the monitoring problem. In terms of performance, the salient quality of FGM is that it can adapt naturally to adverse changes in the monitored problem, such as lack of monotonicity or very tight monitoring bounds, where no method can deliver asymptotically good performance. We provide formal proofs for many of the properties of FGM, and present an extensive empirical performance evaluation under adverse conditions, on real data.

Available Files

Services

Statistics