1. 1 Provenance in O RCHESTRA T.J. Green, G. Karvounarakis, Z. Ives, V. Tannen University of Pennsylvania Principles of Provenance (PrOPr) Philadelphia, PA June 26, 2007

2. 2 Collaborative Data Sharing [Ives+ CIDR 05] (m 1 ) G ( i, c, n ) ! B ( i, n ) (m 2 ) G ( i, c, n ) ! U ( n, c ) (m 3 ) B ( i, n ) !  c U ( n, c ) (m 4 ) B ( i, c )  U ( n, c ) ! B ( i, n ) B(id,nam) G(id,can,nam) m2m2 m4m4 m3m3 m1m1 P GUS P uBio P BioSQL U(nam,can) + - + + - + BB + - + UU + - + UU +U(3,2) comes from +G(1,2,3) via m 2 Schema mappings specify how data is logically related Update exchange propagates updates and records provenance information (1) to assess trust conditions (2) to facilitate incremental maintenance

3. 3 Insertions and provenance +G(3,5,2) p 3 +B(3,5) p 1 +G(1,2,3) p 4 +U(2,5) p 2 (3,5,2) (1,2,3) (3,5) (3,2) (3,3) (1,3) (5,c 1 ) (2,5) (2,c 2 ) (3,c 3 ) (3,2) m1m1 m1m1 m2m2 m3m3 m4m4 m4m4 m3m3 m3m3 m3m3 m2m2 GB U

4. 4 Deletions and provenance +G(3,5,2) p 3 +B(3,5) p 1 +G(1,2,3) p 4 +U(2,5) p 2 GB U (3,5,2) (1,2,3) (3,5) (3,2) (3,3) (1,3) (5,c 1 ) (2,5) (2,c 2 ) (3,c 3 ) (3,2) m1m1 m1m1 m2m2 m3m3 m4m4 m4m4 m3m3 m3m3 m3m3 m2m2

5. 5 Trust and provenance +G(3,5,2) p 3 +B(3,5) p 1 +G(1,2,3) p 4 +U(2,5) p 2 GB U (3,5,2) (1,2,3) (3,5) (3,2) (3,3) (1,3) (5,c 1 ) (2,5) (2,c 2 ) (3,c 3 ) (3,2) m1m1 m1m1 m2m2 m3m3 m4m4 m4m4 m3m3 m3m3 m3m3 m2m2 Peer B distrusts any tuple B(i,n) that came from mapping m4 if n  2. Peer B distrusts any tuple B(i,n), if the data came from Peer G and n ¸ 3 and trusts any tuple from Peer U

6. 6 Further aspects of O RCHESTRA ● Semantics: insertions and deletions with idbs ● Handling conflicts among trusted updates (Taylor+Ives SIGMOD 06) ● Prototype implementation (demo SIGMOD 07, technical details VLDB 07): ▪ Java middleware layer using database as subcomponent ▪ Provenance expressions stored as tables ▪ tgds become datalog rules with Skolem functions ▪ Update exchange using relational query engine (recursion!) ▪ Feasibility experiments ● Future work / topics for discussion ▪ What else to do with rich provenance information? (ranked trust models, bag semantics, querying provenance,...)

7. 7 Trust policies ● Peer B distrusts any tuple B(i,n) that came from mapping m 4 if n  5. ● Peer B distrusts any tuple B(i,n), if the data came from Peer G and n ¸ 3 and trusts any tuple from Peer U