Presentation is loading. Please wait.

Presentation is loading. Please wait.

Efficient Skyline Computation on Vertically Partitioned Datasets Dimitris Papadias, David Yang, Georgios Trimponias CSE Department, HKUST, Hong Kong.

Similar presentations


Presentation on theme: "Efficient Skyline Computation on Vertically Partitioned Datasets Dimitris Papadias, David Yang, Georgios Trimponias CSE Department, HKUST, Hong Kong."— Presentation transcript:

1 Efficient Skyline Computation on Vertically Partitioned Datasets Dimitris Papadias, David Yang, Georgios Trimponias CSE Department, HKUST, Hong Kong

2 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

3 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

4 A Motivating Example  Consider a database containing information about hotels. The y-dimension represents the price of the room, whereas the x-dimension captures the distance of the room from the beach. Distance Price Skyline objects Hotel roomsp Dominance Region of p Borders of p’s Dominance Region

5 Skyline Preliminaries [ICDE, 2001]  Skylines constitute a very useful tool in numerous disciplines, such as for multidimensional decision making and data mining.  Given a set of d-dimensional objects p 1, …, p N, the skyline operator retrieves all these objects that are nor dominated by any other object in the set.  An object p i dominates another point p j, if it is not worse than p j in all dimensions and better than it in at least one dimension.  Properties: The top-1 tuple according to any preference function that assigns scores to tuples is in the skyline tuple. Conversely, for any skyline tuple, there exists a preference function according to which it is the top-1.

6 4 Common Data Distributions

7 Problem Definition  Compute the skyline when the dataset is vertically decomposed among a set of N servers.  Goal: minimize the data that must be retrieved from each server.  We assume wireless environments, where communication overhead constitutes the dominant factor in battery consumption.  Consider mobile phone applications as real world examples.

8 (a) Subspace D 1 at server N 1 (b) Subspace D 2 at Server N 2 First Observations  The global skyline may contain points that do not appear in the local skylines.  Instead of transmitting all records over the network, avoid sending out points that are guaranteed to be dominated globally by an anchor point.

9 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

10 Balke’s Algorithm [EDBT, 2004]  Assume that the d-dimensional database is vertically partitioned into d lists, one for each dimension, assigned to different servers. The lists contain values in ascending order.  Idea: perform sorted accesses on the d lists in a round- robin manner, until a point p (anchor), is reached in every list.  Points that have not showed up at this moment in any list can be safely pruned, since they are dominated by the anchor.

11 Example  Let a 2-dimensional database with the following two lists:  L1  L2 Pointabdmgc Value123556 Pointcdekab Value123457 … …

12 Example (cont.)  Let a 2-dimensional database with the following two lists:  L1  L2 Pointabdmgc Value123556 Pointcdekab Value123457 … … The first point to be retrieved from both lists.

13 Example (cont.)  Let a 2-dimensional database with the following two lists:  L1  L2 Pointabdmgc Value123556 Pointcdekab Value123457 … … The first point to be retrieved from both lists.These points cannot be part of the skyline.

14 Further Improvement  Efficiency can be improved, if instead of visiting the lists in a round-robin manner, we access the most promising list with random accesses.  As a result, only the least expansion is performed on each list. ∙ ∙ ∙ P ∙ ∙ ∙ L1 ∙ ∙ ∙ P ∙ ∙ ∙ L2 avoid visiting these points

15 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

16 Setting  Let N 1,.., N m be m servers storing the same dataset DB.  For each record P  DB every server N i maintains a local score s i (P), and sorts all records in decreasing order of their local scores.  A client wishes to obtain the k records of DB with the maximum global score s.  The score is computed using a monotonic function f on the local scores, i.e., s(P) = f(s 1 (P),.., s m (P)).  Goal: minimize the required number of accesses.

17 Fagin’s Algorithm [PODS, 2001]  Each server N i performs sorted round-robin accesses and sends to the client the next record and its local score.  When the first common record P anc is encountered by all servers, the client terminates the sorted accesses.  Then, it obtains the missing local scores of the other encountered points through random accesses.  The candidate with the highest global score is the top-1 result.

18 Threshold Algorithm [PODS, 2001]  It utilizes an upper bound  TA on the global score to terminate earlier than FA.  The client retrieves the local scores of newly encountered points with random accesses at the remaining servers and computes their global scores, and picks the best score s best.  The threshold  TA is equal to the sum of the local thresholds at each server.  As long  TA > s best, TA continues the sorted accesses, while it keeps updating  TA.  Eventually, the top-1 point will be returned.

19 Example for FA and TA

20 Best Position Algorithm [VLDB, 2007]  It further improves TA by utilizing a tighter threshold.  Let bp i be the position at server N i such that all points up to bp i have been encountered through sorted or random accesses.  The global threshold  BP is equal to the sum of the local thresholds at bp i.

21 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

22 Flajolet / Martin sketches [JCSS ’85]  Goal: Estimate the distinct number of objects from a small-space representation of a set.  Sketch of a union of items is the OR of their sketches  Insertion order and duplicates don’t matter! Prerequisite: Let h be a random, binary hash function. Sketch of an item For each unique item with ID x, For each integer 1 ≤ i ≤ k in turn, Compute h (x, i). Stop when h (x, i) = 1, and set bit i. X 00100 Z 10000 X Z 10100 ∩

23 Flajolet / Martin sketches (cont.) Estimating COUNT Take the sketch of a set of N items. Let j be the position of the leftmost zero in the sketch. j is an estimator of log 2 (0.77 N) Fixable drawbacks: Estimate has faint bias Variance in the estimate is large. 110 1 S 1 Best guess: COUNT ~ 11 j = 3

24 Flajolet / Martin sketches (cont.)  Standard variance reduction methods apply.  Compute m independent sketches in parallel.  Compute m independent estimates of N.  Take the mean of the estimates.  Provable tradeoffs between m and variance of the estimator.

25 Application to COUNT in Sensor Databases Each sensor computes k independent sketches of itself (using unique ID x) –sensor computes a sketch of its value. Use a robust routing algorithm to route sketches up to the sink. Aggregate the k sketches via union en-route. (OR) The sink then estimates the count. sink S1S1 S3S3 S2S2 S4S4 S1S1 S2S2 S1∪S2∪S3S1∪S2∪S3 S4S4 S1∪S2∪S3∪S3S1∪S2∪S3∪S3

26 Outline  Problem Statement  Skyline Computation on Vertically Partitioned Datasets using Balke’s Algorithm  Algorithms for Top-k Query Processing  FM Sketches  Putting Everything Together

27 Problem Characteristics  Each vertical decomposition has arbitrary dimensionality, contrary to Balke’s setting.  Anchor selection substantially determines the total number of transmitted data.  VPS adopts sorting on the local dominance. In particular, the local dominance count dom i (P) of a point P with respect to subspace D i is the number of points dominated by P in D i.  Balke selects as the anchor, the data point P with the maximal dom SUM (P).  We utilize a tighter upper bound for dom(P) is the minimum dom MIN among all local dominance counts.

28 Anchor Selection (a) Subspace D 1 at server N 1 (b) Subspace D 2 at Server N 2 C: optimal anchor point A: has maximal dom MIN B: has maximal dom SUM

29 Our algorithm on the previous example

30 1 st Optimization: Multiple Anchor Points  The previous algorithm performs pruning with a single anchor P anc. Specifically, a point P that is locally dominated by P anc in all subspaces is not sent to the client.  On the other hand, if P is incomparable with P anc even in a single subspace D i, it will be transmitted by the corresponding server N i.  We suggest that multiple points can often achieve more effective pruning.

31 Pruning with 2 points

32 2 nd Optimization: Integration of Sketches  So far, we have estimated the (expected) global dominance dom(P) of a point P using dom MIN (P).  This approach is biased towards points that have high local dominance counts in all subspaces, but dominate few records globally (A).  Thus, we propose an unbiased approach that directly estimates the global dominance counts using sketches that count the number of distinct objects approximately.  We assume that each N i server has a local dominance sketch sk i (P) for every point P, which aggregates all points that P dominates locally in D i.

33 Experiments

34 Thank you!


Download ppt "Efficient Skyline Computation on Vertically Partitioned Datasets Dimitris Papadias, David Yang, Georgios Trimponias CSE Department, HKUST, Hong Kong."

Similar presentations


Ads by Google