Replicating Basic Components

1
Replicating Basic Components

• Bettina Kemme
• McGill University,
• Montreal, Canada

2
Outline

• Previous Work Database replication
• Interest
• Resources
• Current related work

3
Group Communication and Replica Control
DB
Group Comm.
DB

• GC provide
• total order multicast

Serialize transactions according to total order
4
Embedded Architecture
DB
DB
Repl. Tool
Repl. Tool
GC
GC
5
Implemented Features

• A client connects to one site
• But this can be any site in the system
• Enhanced concurrency protocol guarantees
  serializability
• Execution Model
• Communication at the END of the transaction once
  all update operations are determined
• reexecute updates at other sites
• Only apply changes at remote sites
• Failure doesnt stop other sites
• Only clients connected to failed site must
  reconnect
• Semantic for these clients as usual an
  unanswered commit request might have succeeded or
  not

6
Implemented Features

• Online recovery
• One site transfers state to recovering site
• Other sites continue execution
• Simple transfer scheme (entire database)
• Only basic handling in case new failure occurs
  during online recovery
• Partial Replication
• Requests send to all
• Filtering at receiving sites
• Automatic redirecting of queries that cannot be
  answered locally

7
Developed Features

• Online recovery
• Minimizing state transfer costs
• Advanced handling of failures during
  reconfiguration
• Exactly-once transactions in failure-recovery
  model
• When a transaction commits at a site shortly
  before failing, the transaction should commit at
  the available sites
• When a site recovers, for a given transaction
• The result of the transaction is already in the
  DB of the recovering site
• It receives the results within the state transfer
• It receives the transaction after finishing the
  transfer
• Current work with Alberto
• Only GC is not enough because transactions
  commits some time after message is delivered --
  failure in between possible

8
Middleware Architecture
Replication Tool
Replication Tool
GC
GC
DB
DB
9
Implemented Features

• Client currently part of the group
• Message is forwarded at BOT
• Concurrency control in middleware layer at BOT
• One site executes transaction
• Message with changes at EOT
• Failure Model
• Execution guarantee once the first message made
  it
• Concept usable for ADAPT middleware layer?

10
Developed Features

• Optimistic Execution
• Overlap communication overhead and transaction
  processing
• Might be of interest in WAN

11
Main Interests

• In basic services
• All components
• Communication -- working with Alberto and Ozalp
• Transactional
• More complex transactional models in basic
  components (J2EE is flat -- how do we want to
  extend?)
• Replication of stateful services
• In middleware architecture
• Intra-organizational and inter-organizational

12
Resources Available

• Funding
• General funding situation regarding IST-project
• Funding for PhD student guaranteed for the entire
  period
• Technical assistance likely to be funded
• Hardware/Software is secured
• Travel

13
Resources Available

• Current situation
• PhD student started on topic in July
• Implemented TPC-W on top of JBoss in order to
  learn J2EE
• Starts to have reasonable background in
  transactions/replication/fault-tolerance/group-com
  munication/object management
• Master students work on related topics
• Hardware/Software
• Lab more or less established
• New infrastructure will be installed within the
  next year

14
Related Work

• Current situation Postgres-R
• One master student works on it
• Currently Migration to PostgreSQL 7.2
• Open Source Involvement
• See Gborg web-page on replication
• Company funding
• Basic Components for Bioinformatics
• Web-based/application server infrastructure
• Access control