1. Replication Chapter 14 1-3 Katherine Dawicki

2. Motivations Performance enhancement Increased availability Fault Tolerance

3. Increased availability Ideal proportion of time service is accessible with reasonable response time is close to 100% Factors –server failure, – network partitions and disconnected operations

4. Fault Tolerance Replication transparency Consistency

5. Figure 14.1 A basic architectural model for the management of replicated data FE Requests and replies C Replica C Service Clients Front ends managers RM FE RM

6. Replica Manager Applies operations to its replicas recoverably Sometimes required to be a state machine Often each replica manager maintains a replica of every object The set of replica managers may be static or dynamic

7. Figure 14.1 A basic architectural model for the management of replicated data FE Requests and replies C Replica C Service Clients Front ends managers RM FE RM

8. Five phases of performing a request on a replicated object Request Coordination Execution Agreement response

9. Figure 14.2 Services provided for process groups Join Group address expansion Multicast communication Group send Fail Group membership management Leave Process group

10. Role of the group membership service Providing an interface for group membership changes Implementing a failure detector Notifying members of group membership changes Performing group address expansion

11. View delivery Requirements –Order –Integrity –Non-triviality

12. View-synchronous group communication Guarantees –Agreement –Integrity –Validity

13. Figure 14.3 View-synchronous group communication

14. Fault-tolerant Services Client1:Client2: setBalanceB(x,1) setBalanceA(y,2) getBalanceA(y) 2 getBalanceB(x) 0

15. Linearizability and sequential consistency Replicated shared Object is lineraizable –If the interleaved sequence of operations meets the specification of a single correct copy of the objects –The order of operations in the interleaving is consistent with the real times at which the operations occurred in the actual execution Replicated shared Object is sequentially consistent –The interleaved sequence of operations meets the specification of a single correct copy of objects –The order of operations interleaving is consistent with the program order in which the individual client executed them

16. sequentially consistent example Client 1:Client 2: setBalanceB(x,1) getBalanceA(y) 0 getBalanceA(x) 0 setBalanceA(y,2)

17. Figure 14.4 The passive (primary- backup) model for fault tolerance FE C C RM Primary Backup RM

18. Sequence of events for client request (Passive Replication) Request Coordination Execution Agreement Response

19. Passive replication implements linearlizability The primary sequences all the operations upon shared objects If primary fails then –The primary is replaced with a single unique backup –The replica managers that survive agree on which operations had been performed at the point when the replacement primary takes over

20. Figure 14.5 Active replication FEC CRM

21. Sequence of events for client request (Active Replication) Request Coordination Execution Agreement -none Response

22. Active replication implements sequential consistency The reliability of the multicast ensures that every correct replica manager process the same set of requests The total order ensures that they process them in the same order. Each front end’s request are served in FIFO order which is the same as program order.

23. Exercise 14.2 Explain why a multi-threaded server might not qualify as a state machine.

24. Exercise 14.4 A router separating process p from two others, q and r, fails immediately after p initiates the multicasting of message m. If the group communication system is view- synchronous, explain what happens to p next.