ML Wiki

Replication

Replication is a process of updating the same information stored in Distributed Databases

• so all servers have the up-to-date information

Strategies

• synchronized copies
  • costly, often unnecessary
  • use Two-Phase Commit or similar strategies
  • i.e. need asynchronous techniques for this

Asynchronous Replication

Asymmetric Replication

Main idea:

• one primary copy where all changes are performed
• several secondary copies that do not accept writes - only reads
• secondary copies are updated asynchronously with the primary one

Modules

• capture module: monitors changes made to the primary copy
• application module: propagates changes to the secondary copies

Symmetric Replication

• all copies accept updates
• each can have both capture and application modules
• but simultaneous updates may cause loss of consistency
  • i.e. the copies may be inconsistent during some period of time
• this leans to another notion of consistency: Eventual Consistency

Examples

• withdrawal in an ATM
  • not instant - it takes some time for changes to propagate to every database
  • so it's not consistent during some period of time until changes are eventually propagated
  • 
• suppose we have a network
  • and in for one server the net goes down
  • this branch still continue working with database: updates are performed locally
  • when the network is restored, the changes are propagated
  • 

Ways of Replicating

Replication with Triggers

In Active Databases this can be done with triggers

• so we accumulate positive and negative changes into PosDelta and NegDelta
• then these deltas are applied to the secondary copies

CREATE RULE Capture1 ON Primary
WHEN INSERTED
THEN INSERT INTO PosDelta (SELECT * FROM INSERTED)

CREATE RULE Capture2 ON Primary
WHEN DELETED
THEN INSERT INTO NegDelta (SELECT * FROM DELETED)

CREATE RULE Capture3 ON Primary
WHEN UPDATED
THEN INSERT INTO PosDelta (SELECT * FROM INSERTED);
	INSERT INTO NegDelta (SELECT * FROM DELETED);

Replication with Logs

Redo Logging or Undo/Redo Logging mechanisms can be used to do that

• keep track on all committed changes
• eventually apply them to the secondary copies

Sources

• Advanced Databases (ULB)