Self-managing load shedding for data stream management systems
DocUID: 2013-001 Full Text:
PDF Author: Thao N. Pham, Panos K. Chrysanthis, Alexandros Labrinidis
Abstract: Load shedding is an integral component in many Data Stream Management Systems, aiming at preventing the response time from exceeding a user-specified delay target under overload situations. The currently best performing load shedder determines the correct amount of load to shed by utilizing a feedback loop for correcting the statistics-based estimations. Although this load shedder outperforms previous works in controlling response time as well as minimizing data loss, it requires a manually-tuned parameter and cannot work with complex query networks containing joins, aggregations or shared operators. In this paper, we propose SEaMLeSS - SElf Managing Load Shedding for data Stream management systems, which extends and rectifies these limitations of the state-of-the-art load shedder while making it applicable for multi-tenant servers. We implement and evaluate our extensions in AQSIOS, our experimental DSMS prototype, using both synthetic and real input patterns.
Published In: The Eighth International Workshop on Self-Managing Database Systems
Pages: pp. 1-7
Year Published: 2013
Project: AQSIOS Subject Area: Data Streams
Publication Type: Workshop Paper
Sponsor: NSF IIS-0534531, NSF CAREER IIS-0746696