Survey of PeertoPeer Database

1
Survey of Peer-to-Peer Database

• Ming Zhang Yang Li

2
Outline

• Introduction
• Review Current Research PDBMS
• Issues of PDBMS
• Discussion
• Existing PDBMS projects Piazza, Hyperion, PeerDB

3
Introduction

• Peer-to-Peer Database Management Systems (PDBMS)
• P2P systems becomes popular
• Most P2P system lacks data management system
• Database in P2P system is attractive.
• Mobile users can share the same functionalities
• The research is still in its beginning of
  evolution.

4
Review and Discussion of PDBMS

• P2P System vs. Distributed Database System
• A Model of PDBMS

5
PDBMS vs. Distributed Database System

• PDBMS nodes can join or leave the network at any
  time.
• Distributed Database System nodes may join or
  leave the network in a control manner, i.e., add
  when needed

6
PDBMS vs. Distributed Database System

• PDBMS schema is not global.
• Distributed Database System nodes are usually
  stable and standard, and have some knowledge of a
  shared schema

7
PDBMS vs. Distributed Database System

• PDBMS nodes may not contain the complete data.
  Nodes may not be connected
• Distributed Database System containing a
  complete set of data in each server cluster

8
PDBMS vs. Distributed Database System

• PDBMS queries must be routed to many nodes in
  order to return an accurate result set
• Distributed Database System a query can be
  routed to a relatively small set of nodes

9
A Model of PDBMS

• Requirements
• Coordinating databases
• Communicate freely in decentralized environment
• Autonomous Peers data, schema, choice of peers

10
A Model of PDBMS

• Local Relational Model (LRM)
• Assumptions
• The set of all data consists of local
  (relational) databases
• With a set of acquaintances (peers)
• Peers are fully autonomous in choosing their
  acquaintances
• Join or leave the network at any time

11
A Model of PDBMS

• Local Relational Model (LRM)
• Main Goals
• Allow for inconsistent databases
• Support semantic interoperability
• Drawbacks
• Need a protocol to establish acquaintance
  dynamically
• Meta data management?
• Need a query propagation mechanism

12
A Model of PDBMS
Architecture of an LRM node
13
PDBMS Strengths

• No single point of failure
• Data high distributed by using local caching
• If peer fails, query can still continue
• Minimal administration
• Distributed DB require extensive administration
• P2P users create their own database and perform
  their own database manipulation

14
PDBMS Strengths

• Vast amount of data
• User access and search a very large amount of
  data with many different types
• Replicated-data for fast retrieve
• Data is replicated over many nodes as queries are
  performed
• The results can be cached locally
• Fast access local data if the original nodes are
  busy

15
PDBMS Weaknesses

• Discovery of peers
• Peer nodes join and leave freely
• Query routing
• Queries must be routed to all or large subnets of
  nodes
• Consumption of network resources
• Because of intensive routing
• Mobility of users
• Nodes are assigned dynamic IP, so mapping table
  must be maintained and read before queries being
  routed.

16
Discussion

• Key Issues
• Scalability
• Scale in a less than linear fashion
• Availability
• A has, B caches, C accesses from A and/or B
• Performance
• Data Authenticity
• Distinguish a correct/incorrect query
• Security
• Authorize users to access privileged data

17
Case Study

• 1. The Piazza Peer Data Management Project
• 2. The Hyperion Project From Data Integration to
  Data Coordination
• 3. PeerDB A P2P-based System for Distributed
  Data Sharing

18
The Piazza Peer Data Management Project

• University of Washington
• The Piazza focuses on the problem of sharing
  semantically heterogeneous data in a distributed
  and scalable way.

19
Piazza Cont.

• Schema Mediation
• Querying
• Constructing Mappings

20
Piazza Cont.

• Schema Mediation

21
Piazza Cont.

• Querying
• Unfolding replaces a subgoal with a set of
  subgoals.1
• Rewriting replaces a set of subgoals with a
  single subgoal.2
• Constructing Mappings
• Schema Matching
• Precise Mapping

22
The Hyperion Project From Data Integration to
Data Coordination

• University of Toronto
• The Hyperion focuses on the specification and
  management of the logical meta-data that enables
  data sharing and coordination between
  independent, autonomous peers.

23
Hyperion Cont.

• Infrastructure
• Querying
• Coordination

24
Hyperion Cont.

• Infrastructure

25
Hyperion Cont.

• Querying
• Hyperion assumes each query is defined with
  respect to the schema of a single peer.
• A local query is executed using only the data in
  the local peer, while a global query uses data in
  other peers.
• Use any available mapping expressions and mapping
  tables to translate or rewrite.
• Coordination
• Use event-condition-action (ECA) rules to
  coordinate between peers

26
PeerDB A P2P-based System for Distributed Data
Sharing

• National University of Singapore
• PeerDB is a prototype P2P distributed object
  management system that incorporates some features
  such as full fledge object management, without a
  shared global schema and mobile agents.

27
PeerDB Cont.

• Architecture of a PeerDB Node
• Agent Assisted Query Processing
• Cache Management

28
PeerDB Cont.

• Architecture of a PeerDB Node

29
PeerDB Cont.

• Agent Assisted Query Processing
• In the first phase, PeerDB apply the relation
  matching strategy to locate potential relations.
• In phase two, PeerDB sends the queries directly
  to the nodes containing the selected relations.
• Finally the answers are returned to the query
  node.

30
PeerDB Cont.

• Cache Management
• caches answers returned from remote nodes
• periodically invalidates to keep them up-to-date
• LRU (least recently used )replacement policy

31
References

• 1 J. Ullman. Database and Knowledge-Base
  Systems, volume 2. Addison-Wesley, 1989.
• 2 A. Halevy. Answering queries using views a
  survey. VLDB Journal,10(4), 2001.
• 3 Wee Siong Ng, Beng Chin Ooi, PeerDB A
  P2P-based System for Distributed Data Sharing,
  http//www.cs.cornell.edu/courses/cs732/2003sp/pap
  ers/Ng2003.pdf

32
References

• 4 W Anthony Young, Evaluation of Peer-to-Peer
  Database Solutions, http//www.tonyyoung.ca/cs654p
  aper.pdf
• 5 Albena Roshelova, A Peer-to-Peer Database
  Management System, http//eprints.biblio
  .unitn.it/archive/00000585/01/albena.pdf
• 6 I.Tatarinov, Z. Ives, J. Madhavan, A. Halevy,
  D. Suciu, N. Dalvi, X. Dong, Y.Kadiyska, G.
  Miklau, and P.Mork The Piazza Peer Data
  Management Project. SIGMOD Record, ACM, September
  2003

33
References

• 7 Z. Ives, A. Halevy, and D. Weld. Integrating
  Network-Bound XML Data. IEEE Data Engineering
  Bulletin, 24(2), 2001.
• 8M. Arenas, V. Kantere, A. Kementsietsidis, I.
  Kiringa, R. J. Miller, J. Mylopoulos The
  Hyperion project From Data Integration to Data
  Coordination. University of Toronto, Canada,
  2003.
• 9W.Ng, B. Ooi, K.Tan, and A. Zhou. PeerDB A
  p2p-based system for distributed data sharing.
  International Conference on Data Engineering
  (ICDE), 2003.