Click here to load reader
Upload
amplephi
View
499
Download
1
Embed Size (px)
Citation preview
SMM: A Data Stream Management System for Knowledge Discovery
1
Hetal Thakkar, Nikolay Laptev, Hamid Mousavi, Barzan Mozafari, Vincenzo Russo, Carlo Zaniolo
Computer Science Department UCLA
Data Stream Management Systems (DSMS)
2
• DSMS critical in a variety of applicationso Click-stream analysis,o Algorithmic Tradingo Network monitoringo Credit card fraud detection …
• Many DSMS Projects and Prototypes :o STREAM (Stanford), Aurora/Borealis (Brown, MIT),
Telegraph (UCB), Gigascope (AT&T), Stream Mill (UCLA), … and so on.
• Commercial Startups and vendor extensions:o StreamBase, Aleri, Coral8, Apama, Truviso,o DBMS vendors …
• Support for online mining on data streams: unresolved issue for current systems.
Two Main Research Challenges
3
• Challenge I: Fast and Light algorithms needed for online mining algorithms.
• Challenge II: These and business intelligence applications require the Quality of Service (QoS) of DSMS. Thus these algorithms must be deployed as part of a DSMS.
• Much research on first challenge—a stream of papers in DM conferences—but not on the second that is probably even harder.
Data Stream Mining & DSMS QoS
4
• DSMS: Support continuous queries over massive data streams – with QoS (Quality of Service) guarantees and– (Quasi) Real-time response through:
o Scheduling, query optimization,o Windows and other Synopseso Load shedding …
• But - Current DSMS focus on simple continuous queries- Using query languages based on SQL- Lackluster history of SQL with KDD- DSMS bring more problems:
e.g. blocking queries not allowed.