Data Replication

Outline
 Introduction  Distributed DBMS Architecture  Distributed Database Design  Distributed Query Processing  Distributed Transaction Management  Data Replication
 Consistency criteria  Update propagation protocols

 Parallel Database Systems  Data Integration Systems  Web Search/Querying  Peer-to-Peer Data Management  Data Stream Management
Distributed & Parallel DBMS M. Tamer Özsu Page 6.1

Acknowledgements
 Many of these slides are from notes prepared by  Prof. Gustavo Alonso, ETH Zurich, Switzerland  Prof. Bettina Kemme, McGill university, Canada

Distributed & Parallel DBMS

M. Tamer Özsu

Page 6.2

Replication
 Why replicate?
 Reliability
 

Avoid single points of failure Scalability in numbers and geographic area

 Performance

 Why not replicate?
 Replication transparency  Consistency issues
Updates are costly  Availability may suffer if not careful


Distributed & Parallel DBMS

M. Tamer Özsu

Page 6.3

Execution Model
 There are physical copies of logical objects in the system.  Operations are specified on logical objects, but translated to

operate on physical objects.

Write(x) x Logical object

Write(x1) Write(x2) Write(xn)

x1

x2

…

xn

Physical copies
Distributed & Parallel DBMS M. Tamer Özsu Page 6.4

Replication Issues
 Consistency models - how do we reason about the consistency of the “global state”?
 Transactional criteria
One-copy serializability  Snapshot isolation  Strong session serializability
  

 Object-centric consistency  Bounded inconsistency
Eventual consistency Epsilon serializability

 Update propogation protocols - what is the algorithm that ensures mutual consistency of replicas by enforcing a given consistency model?

Distributed & Parallel DBMS

M. Tamer Özsu

Page 6.5

Transactional Replica Consistency
 Efficient implementation of sequential consistency requires transactions.  One-copy equivalence
 The