Presentation is loading. Please wait.

Presentation is loading. Please wait.

Prefer: A System for the Efficient Execution

Similar presentations


Presentation on theme: "Prefer: A System for the Efficient Execution"— Presentation transcript:

1 Prefer: A System for the Efficient Execution
of Multi-parametric Ranked Queries Vagelis Hristidis Nick Koudas Uannis Papakonstantinou Presented By: Amrita Tamrakar 09-Mar-2006

2 Contents Introduction Importance/Weights of the attributes
The Efficiency and response time ?? Solution View Selection Pipelining Algorithm Conclusion

3 Prefer Prefer is a layer on top of commercial database which allows efficient evaluation of multi-parametric databases. Different attributes may have diff weight House price and bedrooms have more weight than the age. Cheap houses may be old and small (so search more)

4 Importance of Attributes
Preference is the keyword Weights a1 a2 a3 …. an Preference function a1A1+a2A2 +….+Anan Attributes A1 A2 A3 …. An : tuples

5 Prefer Interface

6 Example Tid A1 A2 A3 Score 1 10 17 20 17.2 2 11 17.3 Preference = (0.1,0.6,0.3) , Score 10*0.1+17*0.6+20*0.3

7 Efficiency and Response Time
Retrieve the whole Table Apply the preference function on each tuple Sort the tuples Prf fn Relation Score Sort Top k tuples ……….……….

8 Solution Proposed Ranked Materialized view
10-20 views can have most possible queries Queries with linear preference function ∑ Preference vector v = {v1,....vn} Attribute preference

9 Prefer Architechture Best View for the query Materialized Views
Preprocessing Stage Select Materialized View Best View for the query Materialized Views Execute Pipelining Algorithm Output Results Finding the top k and sorting View Selection

10 View Selection The union of views will cover the whole space
Outputs view sequence V such that for every query q there is at least one view that covers q Greedy view selection algorithm for space constraints Given a set of views { Rv1….Rvn } that covers [0,1]k space, select C views that maximize the number of points in [0,1]k covered.

11 Pipelining Ranked query using
ranked view T = DeteremineWatermark() Scan Rv for tuples greater than T Sort these tuples by fq and mark as processed Continue for unprocessed tuples till top –k is retrieved v R v Rv1 Ranked tuples Query q tqi tq1 R q

12 Example Tid A1 A2 A3 fv(t) fq(t) 1 10 17 20 16.8 17.2 2 11 16.4 17.3 3
Q=(0.1,0.6,0.3) v=(0.2,0.4,0.4) Tid A1 A2 A3 fv(t) fq(t) 1 10 17 20 16.8 17.2 2 11 16.4 17.3 3 18 12 15.4 16.1 4 15 8 9.8 10.1 5 9 6 Rv1 The first watermark= 14.26 Sort by fq Till k tuples are retrieved

13

14 Conclusion Introduction of database selection queries with user preferences Use of multiple DB views Algorithms to select the best view Practical application via PREFER


Download ppt "Prefer: A System for the Efficient Execution"

Similar presentations


Ads by Google