Explain why, in step (b), all primaries must be updated in a single transaction and, in step (c), the update is propagated to site A.

Question

The following variant of the primary copy asynchronous-update replication protocol has been proposed for totally replicated systems.
a. A transaction executing at site A updates only the replicas at A before it commits
(it needed to access no other site since replication is total).
b. After the transaction commits, a second transaction is initiated to update the primary copies of all items updated at A.
c. After the transaction in step (b) has completed, each primary site propagates the update made at that site to all secondaries (including the one at site A). Updates made to a primary copy by several transactions are propagated to secondaries in the order in which the primary was updated.

What will be an ideal response?

Answer 1

The goal of the protocol is mutual consistency of all replicas. All primaries must be updated in a single transaction to make sure that updates to all primaries made by concurrent transactions occur in the same order. The update must be propagated back
to A to make sure that secondary copies at A are updated in the same order as the primaries. For example, suppose sites A and B both have secondary replicas of x and transactions T1 and T2 executing at A and B respectively update x concurrently. If in step (c) the primary site did not send updates to all sites, then when the protocol completed the value of x at A would be the value written by T2 and at B the value written by T1.