� The transaction is complete after the update is applied
locally; the update is then made to the remote copy
� The remote copy may not include all transactions
� Link failure has no impact on running applications
(1) Write (3) Write
(2) ACK (4) ACK
With asynchronous replication we issue a write to the primary, the primary
acknowledges, we now move on. At some point in time later that write goes to
the secondary, and we acknowledge.
If we are doing this in the subsystem it will track the outstanding writes.
If you are doing asynchronous mirroring at the host, it will write to the primary,
and it will write again to the secondary. When it gets the acknowledgement
back from the primary, the application moves on but the host still holds the
second write whilst it waits for that acknowledgement. It knows how many
I/Os it’s going to be out of step.
With asynchronous mirroring the secondary will be behind. If the primary failed
and you were using asynchronous replication, some data can/will be lost. You
cannot guarantee data consistency.
Whereas with synchronous replication you can guarantee data synchronicity
between the two subsystems.