Scalable delivery of stream query result.ppt

Yongluan Zhou University of Southern Denmark Ali Salehi , Karl Aberer EPFL, Switzerland - Manisha Mishra ? Continuous queries over data streams typically produce large volumes of result streams whose delivery from the processing server to the end users is bandwidth consuming and hence should be carefully handled. ? Existing systems often assume that the result streams are directly sent to the users from the server . Such an architecture cannot scale to a large population of users. Naive Approach ? One query one query result stream . ? One query result stream unique identifier . ? User's subscription unique identifier . (. user's data interest) but Reason?? Inefficient Large communication over head overlap contents Q1 SELECT S2. * FROM Station1 [Range 30 Minutes] S1, Station2 [Now] S2 WHERE > Q2 SELECT , , , FROM Station1 [Range 1 Hour] S1, Station2 [Now] S2 WHERE > Q3 SELECT S2. * , , FROM Station1 [Range 1 Hour] S1, Station2 [Now] S2 WHERE > n1 n2 n3 n4 Q1 Q2 S1 S2 S2 S1 Non-Share n1 = Responsible for processing Q1 and Q2 n2 = Neighboring Broker of n1 n3 = Post query Q1 n4 = Post query Q2 Overlapping contents of S1 and S2 are transmitted twice over the link between n1 and n2 . Thus S1,S2,S3 = Result stream of Q1, Q2 and Q3 respectively n1 n2 n3 n4 Q1 Q2 S3 S2 S1 Share Query Reformulation Approach ? Relatively Simple .? Easy to be implemented. ? Implemented as middleware between system processing engine and DPSS(more on this later). ? For a group of queries that have overlapping results, the poses a new query Q called Representative Query that contains all the queries in its group. S3 {S1,S2} We send one result stream S3 to n2, and split S3 into two separate streams S1 and S2 at node n2. ? The whole system consists of a stre