Concurrency Control in Distributed Database Systems with Read-Only replication

1
Concurrency Control in Distributed Database
Systems with Read-Only replication

• Anne Denton and
• Vijayakumar Shanmugasundaram

2
Distributed Systems

• Multiple databases that behave as one
• Replication
• Speeds up read
• Slows down write
• Problem compromises local loads
• Can we restore autonomy of databases?

3
Idea

• Databases have r/w section with local locking
• Data from other databases is replicated in a
  read-only section
• Read-only data is validated after all reads and
  writes and before commit
• If validation fails gt abort
• gt optimistic treatment of replicated data

4
How validate?

• Request validation only of items to be read?
• Rather send validation information for entire
  replicated section (similar to request vector)
• Compression? gt P-trees

5
Concurrency Control

• Define new operations and check serializability
  with existing r, w
• Validate
• Reload
• Read
• Read-only transactions dont require validation,
  if multi-version CC and virtual commit-time is
  used

6
Design Choices

• Validate also at beginning of transaction?
• Download invalidated data automatically?
• Effect on read-write transactions
• Effect on read-only transactions
• Regular validity checks?

7
Simulation

• Using OOD
• Java
• JCL and Our own Classes
• Implementation of P-tree?
• Accommodation of any design choices
• Results Analysis University Statistician