1. Presented by Jiwen Sun, Lihui Zhao 24/3/2004
4/25/2017
Schema Mediation in Peer Data Management Systems (Alon Y. Halevy, Zachary G. Ives, Dan Suciu, Igor Tatarinov )
Presented by Jiwen Sun, Lihui Zhao
24/3/2004

2. Introduction Why Peer to Peer Integration with Semantics
4/25/2017
Introduction
Why Peer to Peer
Integration with Semantics
Flexibility, extensible
The paper’s Contribution
Piazza Project
A peer mapping language
Algorithm for query answering

3. Introduction Traditional Integration Formalisms Global as View (GAV)
Mediated schema as views over data sources
Local as View (LAV)
Data sources as views of mediated schema
GAV:
T :- S1, S2, S3
LAV:
S1  T
Med. Schema T S1 S2 S3

4. Introduction GAV and LAV in Piazza (P2P) environment
Define semantic relations locally
Answer queries globally

5. Introduction Properties of a peer Peer Peer Relations Stored Relations
Peer Description
Peer Description
Storage Description
Peer
Peer Relations
Stored Relations

6. Introduction
Emergency Response Example

7. Problem Definition PPL – Peer-Programming Language Storage Description
4/25/2017
Problem Definition
PPL – Peer-Programming Language
Storage Description
- Mappings between stored relations and peer relations
A : R  Q
Peer Mapping
- Mappings between peer relations
Inclusion: Q1(A1)  Q2(A2)
Definitional: P(x) :- P1(x)

8. Problem Definition GAV-like Definition – Definition in Datalog
9DC : SkilledPerson(PID, “Doctor”) : -
H : Doctor(SID, h, l, s, e)
9DC : SkilledPerson(PID, “EMT”) : -
H : EMT(SID, h, vid, s, e)
FS : Schedule(PID, vid),
FS : FirstResponse(vid, s, l, d),
FS : Skills(PID, “medical”)

9. Problem Definition LAV-like Definition – Inclusion Definition
LH : CritBed(bed, hosp, room, PID, status) 
H : CritBed(bed, hosp, room),
H : Patient(PID, bed, status)
LH : EmergBed(bed, hosp, room, PID, status) 
H : EmergBed(bed, hosp, room),

10. Problem Definition Query Answering in a PDMS
A peer answers a query with local stored data
Reformulate the query and forward to neighbour peers
Query is answered by chaining of mapped peers
Mappings is expanded with a rule-goal tree

11. Complexity of Query Answering
Restrictions on peer mappings decides complexity
In general, finding all certain answers is undecidable.
Acyclic Peer Mappings
Only Inclusion Mappings are used => polynomial time
Cyclic Peer Mappings
Replication type cycle only => polynomial time
Comparison Predicates helps reduce complexity

12. Query Reformulation Algorithm
Algorithm Overview
Building a rule-goal tree
Expand tree by combining and interleaving GAV and LAV
Leaves in Storage Description forms are the query results
Tree size may be huge

13. Building a rule-goal tree
1. Make the query root of tree
Q(f1,f2) q
2. Find views cover the query, expand the query use the views
Q(f1,f2) :- SameEngine(f1,f2,e),Skill(f1,s),Skill(f2,s)
Q(f1,f2) q
SameEngine(f1,f2,e)
Skill(f1,s)
Skill(f2,s)

14. Building a rule-goal tree
3. Mappings between peer schemas
r0: SameEngine(f1, f2, e) :- AssignedTo(f1,e), AssignedTo(f2,e)
r1: SameSkill(f1, f2)  Skill(f1,s), Skill(f2,s)
Q(f1,f2) q
SameEngine(f1,f2,e)
Skill(f1,s)
Skill(f2,s) r0
AssignedTo(f1,e)
AssignedTo(f2,e) r1
SameSkill(f1,f2)
SameSkill(f2,f1)

15. Building a rule-goal tree
4. Repeat until all leaves are storage relations
SamEngine(f1,f2,e)
Skill(f1,s)
Skill(f2,s)
Q(f1,f2) q r0 r1
AssignedTo(f1,e)
AssignedTo(f2,e)
SameSkill(f1,f2)
SameSkill(f2,f1) r2 r3
S1(f1,e,_)
S1(f2,e,_)
S2(f2,f1)
S2(f1,f2)
Reformulated query:
Q’(f1,f2) :- S1(f1,e,_), S1(f2,e,_), S2(f1,f2)  S1(f1,e,_), S1(f2,e,_), S2(f2,f1)

16. Query Reformulation Algorithm
Optimizations
Techniques for Pruning Rule-goal Tree Branches
Memorization of nodes
Constraint on nodes, which contradict query
Redundancy detection
Maximizing the techniques
Order for building tree is important
Prioritize node in Piazza system

17. Experiment Bottleneck is finding rewritings from tree
Tree depth matters, not number of nodes

18. Related Work Answering Queries Using Views (AQUV)
Answering queries using views [Halevy]
Minicon: A scalable algorithm for answering queries using views [Pottinger & Halevy]
PDMS vs. Database Federation
DB federation – mapping between stored relations
Loose relationship => scales better
Peers can play different roles
Chaining through peer mappings to locate data

19. Summary PDMS is superior over data integration systems
Ad-hoc, scalable
Decentralized
PPL describes mappings using GAV/LAV
A query reformulation algorithm produces practical results

20. Thank You!